arXiv:1503.08773v2 [astro-ph.GA] 8 Apr 2015
April 9, 2015
Planck intermediate results. XXVII. High-redshift infrared galaxy
overdensity candidates and lensed sources discovered by
Planck
and confirmed by
Herschel-SPIRE
⋆
Planck Collaboration: N. Aghanim54, B. Altieri35, M. Arnaud66, M. Ashdown63,6, J. Aumont54, C. Baccigalupi79, A. J. Banday88,10, R. B. Barreiro59, N. Bartolo27,60, E. Battaner90,91, A. Beelen54, K. Benabed55,87, A. Benoit-L´evy21,55,87, J.-P. Bernard88,10, M. Bersanelli30,47,
M. Bethermin66, P. Bielewicz88,10,79, L. Bonavera59, J. R. Bond9, J. Borrill13,83, F. R. Bouchet55,81, F. Boulanger54,78, C. Burigana46,28,48, E. Calabrese86, R. Canameras54, J.-F. Cardoso67,1,55, A. Catalano68,65, A. Chamballu66,14,54, R.-R. Chary52, H. C. Chiang24,7, P. R. Christensen75,33,
D. L. Clements51, S. Colombi55,87, F. Couchot64, B. P. Crill61,76, A. Curto6,59, L. Danese79, K. Dassas54, R. D. Davies62, R. J. Davis62, P. de Bernardis29, A. de Rosa46, G. de Zotti43,79, J. Delabrouille1, J. M. Diego59, H. Dole54,53, S. Donzelli47, O. Dor´e61,11, M. Douspis54, A. Ducout55,51,
X. Dupac36, G. Efstathiou57, F. Elsner21,55,87, T. A. Enßlin72, E. Falgarone65, I. Flores-Cacho10,88, O. Forni88,10, M. Frailis45, A. A. Fraisse24, E. Franceschi46, A. Frejsel75, B. Frye85, S. Galeotta45, S. Galli55, K. Ganga1, M. Giard88,10, E. Gjerløw58, J. Gonz´alez-Nuevo59,79, K. M. G´orski61,92, A. Gregorio31,45,50, A. Gruppuso46, D. Gu´ery54, F. K. Hansen58, D. Hanson73,61,9, D. L. Harrison57,63, G. Helou11, C. Hern´andez-Monteagudo12,72, S. R. Hildebrandt61,11, E. Hivon55,87, M. Hobson6, W. A. Holmes61, W. Hovest72, K. M. Huffenberger22,
G. Hurier54, A. H. Jaffe51, T. R. Jaffe88,10, E. Keih¨anen23, R. Keskitalo13, T. S. Kisner70, R. Kneissl34,8, J. Knoche72, M. Kunz16,54,2, H. Kurki-Suonio23,41, G. Lagache5,54, J.-M. Lamarre65, A. Lasenby6,63, M. Lattanzi28, C. R. Lawrence61, E. Le Floc’h66, R. Leonardi36, F. Levrier65, M. Liguori27,60, P. B. Lilje58, M. Linden-Vørnle15, M. L´opez-Caniego36,59, P. M. Lubin25, J. F. Mac´ıas-P´erez68, T. MacKenzie20,
B. Maffei62, N. Mandolesi46,4,28, M. Maris45, P. G. Martin9, C. Martinache54, E. Mart´ınez-Gonz´alez59, S. Masi29, S. Matarrese27,60,39, P. Mazzotta32, A. Melchiorri29,49, A. Mennella30,47, M. Migliaccio57,63, A. Moneti55, L. Montier88,10, G. Morgante46, D. Mortlock51, D. Munshi80,
J. A. Murphy74, P. Natoli28,3,46, M. Negrello43, N. P. H. Nesvadba54, D. Novikov71, I. Novikov75,71, A. Omont55, L. Pagano29,49, F. Pajot54, F. Pasian45, O. Perdereau64, L. Perotto68, F. Perrotta79, V. Pettorino40, F. Piacentini29, M. Piat1, S. Plaszczynski64, E. Pointecouteau88,10, G. Polenta3,44, L. Popa56, G. W. Pratt66, S. Prunet55,87, J.-L. Puget54, J. P. Rachen19,72, W. T. Reach89, M. Reinecke72, M. Remazeilles62,54,1, C. Renault68, I. Ristorcelli88,10, G. Rocha61,11, G. Roudier1,65,61, B. Rusholme52, M. Sandri46, D. Santos68, G. Savini77, D. Scott20, L. D. Spencer80, V. Stolyarov6,63,84, R. Sunyaev72,82, D. Sutton57,63, J.-F. Sygnet55, J. A. Tauber37, L. Terenzi38,46, L. Toffolatti17,59,46, M. Tomasi30,47, M. Tristram64,
M. Tucci16, G. Umana42, L. Valenziano46, J. Valiviita23,41, I. Valtchanov35, B. Van Tent69, J. D. Vieira11,18, P. Vielva59, L. A. Wade61, B. D. Wandelt55,87,26, I. K. Wehus61, N. Welikala86, A. Zacchei45, and A. Zonca25
(Affiliations can be found after the references)
Submitted 11 Aug 2014 / Accepted 13 March 2015
Abstract
We have used the Planck all-sky submillimetre and millimetre maps to search for rare sources distinguished by extreme brightness, a few hundred millijanskies, and their potential for being situated at high redshift. These “cold” Planck sources, selected using the High Frequency Instrument (HFI) directly from the maps and from the Planck Catalogue of Compact Sources (PCCS), all satisfy the criterion of having their rest-frame far-infrared peak redshifted to the frequency range 353 – 857 GHz. This colour-selection favours galaxies in the redshift range z = 2–4, which we consider as cold peaks in the cosmic infrared background. With a 4.′5 beam at the four highest frequencies, our sample is expected to include overdensities of galaxies in groups or clusters, lensed galaxies, and chance line-of-sight projections. We perform a dedicated Herschel-SPIRE follow-up of 234 such Planck targets, finding a significant excess of red 350 and 500 µm sources, in comparison to reference SPIRE fields. About 94 % of the SPIRE sources in the Planck fields are consistent with being overdensities of galaxies peaking at 350 µm, with 3 % peaking at 500 µm, and none peaking at 250 µm. About 3 % are candidate lensed systems, all 12 of which have secure spectroscopic confirmations, placing them at redshifts z > 2.2. Only four targets are Galactic cirrus, yielding a success rate in our search strategy for identifying extragalactic sources within the
Planck beam of better than 98 %. The galaxy overdensities are detected with high significance, half of the sample showing statistical significance
above 10 σ. The SPIRE photometric redshifts of galaxies in overdensities suggest a peak at z ≃ 2, assuming a single common dust temperature for the sources of Td = 35 K. Under this assumption, we derive an infrared (IR) luminosity for each SPIRE source of about 4 × 1012L
⊙, yielding star formation rates of typically 700 M⊙yr−1. If the observed overdensities are actual gravitationally-bound structures, the total IR luminosity of all their SPIRE-detected sources peaks at 4 × 1013L
⊙, leading to total star formation rates of perhaps 7 × 103M⊙yr−1 per overdensity. Taken together, these sources show the signatures of high-z (z > 2) protoclusters of intensively star-forming galaxies. All these observations confirm the uniqueness of our sample compared to reference samples and demonstrate the ability of the all-sky Planck-HFI cold sources to select populations of cosmological and astrophysical interest for structure formation studies.
Key words. Galaxies: high-redshift, clusters, evolution, star formation – Cosmology: observations, large-scale structure of Universe – Submillimetre: galaxies – Gravitational lensing: strong
⋆ corresponding author e-mail: herve.dole@ias.u-psud.fr
1. Introduction
Tegmark et al. 2004, Cole et al. 2005). In the non-linear regime the situation is less clear because structure formation becomes a complex interplay between dark matter collapse and the hydro-dynamics of baryonic cooling. In the particularly dense environs of the most massive dark matter halos, this interplay should lead to vigorous episodes of rapid star formation and galaxy growth, giving rise to copious amounts of FIR (far-infrared) emission from dust heated by young stellar populations in massive galax-ies during periods of intense star formation.
From the standpoint of galaxy evolution, studying this intense star formation epoch in massive dark matter halos may provide a wealth of observational constraints on the kinematics and evolutionary history of galaxies in massive galaxy clusters, as well as the intracluster gas. From the point of view of cosmology, clusters yield information on non-Gaussianities of primordial fluctuations and can challenge the ΛCDM (cold dark matter) model (Brodwin et al. 2010; Hutsi 2010; Williamson et al. 2011; Harrison & Coles 2012; Holz & Perlmutter 2012; Waizmann et al. 2012; Trindade et al. 2013), while lensed sources act as probes of dark matter ha-los, and both are probes of cosmological parameters such as
ΩM (matter density today divided by the critical density of the
Universe) and σ8 (the rms fluctuation in total matter – baryons
+ CDM + massive neutrinos – in 8 h−1Mpc spheres today at z = 0) (Planck Collaboration XVI 2014). Furthermore, clus-ters of galaxies are crucial objects that bridge astrophysics and cosmology, sometimes with some tensions, particularly regard-ing the measurement of σ8 (Planck Collaboration XVI 2014;
Planck Collaboration XX 2014; Planck Collaboration I 2015; Planck Collaboration XIII 2015), all of which has led to a de-bate between cluster phenomenology and cosmological physics. The extragalactic sky in the submillimetre (submm) and millimetre (mm) regime has been of considerable scientific interest for over two decades, with the distinct advantage that the steep rise in the redshifted Rayleigh-Jeans tail of the modified blackbody emitted by the warm dust in in-frared galaxies largely compensates for cosmological dim-ming, the “negative k-correction” (Franceschini et al. 1991; Blain & Longair 1993; Guiderdoni et al. 1997). As a conse-quence, the flux density of galaxies depends only weakly on redshift, opening up a particularly interesting window into the high-redshift Universe (typically 2 < z < 6). Constant improvements in bolometer technology have led to impres-sive samples of high-z galaxies being identified with ground-based, balloon and space-borne telescopes (e.g., Hughes et al. 1998; Barger et al. 1998; Chapman et al. 2005; Lagache et al. 2005; Patanchon et al. 2009; Devlin et al. 2009; Chapin et al. 2009; Negrello et al. 2010; Vieira et al. 2010; Oliver et al. 2010; Eales et al. 2010).
Nevertheless, only with the recent advent of wide-field sur-veys with astrophysical and cosmological scope have the sys-tematic searches become efficient enough to identify the bright-est of these objects with flux densities above about 100 mJy at 350 µm, e.g., with Herschel1 (Pilbratt et al. 2010; Eales et al. 2010; Oliver et al. 2010), the South Pole Telescope (Vieira et al. 2010), WISE (Wright et al. 2010; Stanford et al. 2014), and Spitzer (Papovich 2008; Stanford et al. 2012). Such sources are very rare. For example, the surface density of red sources brighter than 300 mJy at 500 µm is 10−2deg−2 for strongly lensed galaxies, 3 × 10−2deg−2 for AGN (active galactic
nu-1 Herschel is an ESA space observatory with science instruments provided by European-led Principal Investigator consortia and with im-portant participation from NASA.
clei, here radio-loud, mostly blazars), and 10−1deg−2 for
late-type galaxies at moderate redshifts (e.g., Negrello et al. 2007, 2010). Other models predict similar trends (Paciga et al. 2009; Lima et al. 2010; Bethermin et al. 2011; Hezaveh et al. 2012). This makes even relatively shallow submm surveys interesting for searches of high-redshift objects, as long as they cover large parts of the sky. Studies of gravitationally lensed galaxies at high redshifts originating from these surveys (Negrello et al. 2010; Combes et al. 2012; Bussmann et al. 2013; Herranz et al. 2013; Rawle et al. 2014; Wardlow et al. 2013) illustrate the scientific potential of such surveys for identifying particularly interesting targets for a subsequent detailed characterization of high-z star formation through multi-wavelength follow-up observations.
The power of wide-field surveys in detecting the rarest ob-jects on the submm sky is even surpassed with genuine all-sky surveys, which systematically and exhaustively probe the brightest objects in their wavelength domain down to their com-pleteness limits. Here we present a search for the rarest, most extreme high-redshift (z &2) candidates on the submillime-tre sky, which was performed with the Planck2 all-sky survey (Planck Collaboration I 2014, 2015). The Planck Catalogue of Compact Sources (PCCS, Planck Collaboration XXVIII 2014) has a completeness limit of about 600 mJy at the highest fre-quencies, which corresponds to LFIR ≃ 5 × 1013L⊙ at z = 2.
With a 5′ beam (Planck Collaboration VII 2014) at the four
highest frequencies (corresponding to a physical distance of about 2.5 Mpc at z = 2), we expect that sources with bona fide colours of high-z galaxies in the Planck-HFI bands are ei-ther strongly gravitationally lensed galaxies, or the combined dust emission of multiple galaxies in a shared vigorously star-forming environment in the high redshift Universe. The lat-ter case is consistent with the result that submm galaxies or ULIRGs (Ultra Luminous Infrared Galaxies) are strongly clus-tered (Blain et al. 2004; Farrah et al. 2006; Magliocchetti et al. 2007; Austermann et al. 2009; Santos et al. 2011; Ivison et al. 2012; Valtchanov et al. 2013; Noble et al. 2013; Clements et al. 2014), and may include massive galaxy clusters during their ma-jor growth phase. It is also possible of course that the sources are chance alignments of multiple high-redshift galaxies projected onto the same line of sight (Negrello et al. 2005, 2007, 2010; Chiang et al. 2013), or of multiple, lower-mass galaxy groups or clusters.
Identifying high-redshift cluster candidates directly by the signatures of their total star formation is a very useful com-plement to the diagnostics used to identify galaxy clusters so far. Most systematic searches today focus on the pri-mary constituents of more evolved, lower-redshift clusters, like their populations of massive, passively evolving galax-ies (’red-sequence galaxgalax-ies’), the hot intracluster medium (ei-ther through X-ray emission or the Sunyaev-Zeldovich ef-fect), or the suspected progenitors of today’s brightest clus-ter galaxies, in particular high-redshift galaxies (Steidel et al. 2000; Brand et al. 2003; Kodama et al. 2007; Scoville et al. 2007; Venemans et al. 2007; Papovich 2008; Daddi et al. 2009; Brodwin et al. 2010; Galametz et al. 2010; Papovich et al. 2010; Capak et al. 2011; Hatch et al. 2011; Kuiper et al. 2011; Ivison et al. 2013; Muzzin et al. 2013; Wylezalek et al. 2013a;
Chiang et al. 2014; Cooke et al. 2014; Cucciati et al. 2014; Mei et al. 2014; Rettura et al. 2014, for instance). With the lim-ited sensitivity and the large beam of Planck, in turn, we effec-tively select the most intensely star-forming Mpc-scale environ-ments in the high-redshift Universe.
Using Planck, we have selected putative high-redshift ob-jects with spectral energy distributions (SEDs) of warm dust, peaking between observed frequencies of 353 and 857 GHz. This has a net effect of selecting sources either peaking at 545 GHz, or to a lesser extent sources having their infrared peak between 353 and 857 GHz. This equates, in principle, to redshifted infrared galaxies at z ∼ 2–4. We refer to these as “cold” sources of the Cosmic Infrared Background (or CIB), which have a red and thus potentially redshifted SED. The CIB (Puget et al. 1996; Hauser et al. 1998; Hauser & Dwek 2001; Dole et al. 2006; Planck Collaboration XXX 2014) is the integrated relic emission in the broad infrared range, typi-cally 8 µm to 1 mm, where the emission reaches a maximum (Dole et al. 2006). Physically, such objects correspond to galaxy and AGN formation and evolutionary processes, and more gen-erally the history of energy production in the post-recombination Universe (e.g., Kashlinsky 2005; Planck Collaboration XVIII 2011). The CIB, as observed in the submillimetre, is consid-ered as a proxy for intense star formation at redshifts z > 1 (Planck Collaboration XXX 2014) as well as for the mass of structures (Planck Collaboration XVIII 2014).
The Planck collaboration has also released the PCCS (Planck Collaboration XXVIII 2014), containing 24381 sources at 857 GHz, with 7531 sources at Galactic latitudes |b| > 30◦, of
which many are of interest for extragalactic studies. The main difference between the cold sources of the CIB and the red sources of the PCCS (both used in this work) can be summarized as a threshold difference: cold CIB sources are detected in the CIB fluctuations with a combined spectral and angular filtering method (Montier et al. 2010), while PCCS sources are detected independently in the frequency maps using an angular filtering method with a higher threshold (Planck Collaboration XXVIII 2014).
This paper presents the observations and analysis of our ex-tensive dedicated Herschel-SPIRE (Griffin et al. 2010) follow-up of 234 Planck sources (either selected from the CIB fluctu-ations or from the PCCS). The paper is structured into seven sections. In Sect. 2, we detail the Planck parent sample and the Herschel observations. Section 3 gives a technical descrip-tion of the algorithms used in the generadescrip-tion of the SPIRE photometry and the catalogue. In Sect. 4 we use statistics to characterize the Herschel observations; in particular we quan-tify the overdensities and the colours of the SPIRE counterparts and propose a classification of either overdensities or lensed candidates. In Sect. 5, we discuss the properties of the lensed source candidates, while in Sect. 6 we focus on the overden-sities and their characterization, including a stacking analy-sis. Conclusions are reported in Sect. 7. The Appendices con-tain information on the SPIRE catalogue generation and the number counts, and a gallery of sample fields. We use the Planck 2013 cosmology (Planck Collaboration XVI 2014, ta-ble 5: Planck+WP+highL+BAO) throughout the paper.
2. Sample selection and observations
2.1. Planck observations and selection
Planck3observed the whole sky at frequencies between 30 and 857 GHz (Planck Collaboration I 2014). We made two different selections to follow-up with Herschel: first, using the maps and looking for cold sources of the CIB; second, using the public catalogue of compact sources (PCCS).
2.1.1. Cold sources of the CIB in thePlanck maps
We make use of the Planck-HFI (High Frequency Instrument, Planck Collaboration VI 2014) data as well as the IRAS/IRIS data (Miville-Deschˆenes & Lagache 2005) at 100 µm. For this purpose we use the cleanest 26 % of the sky (in the Planck 857 GHz map), defined by a minimal cirrus contamination, NHI < 3 × 1020cm−2. The detection and selection algorithm,
based on Montier et al. (2010) can be summarized in the fol-lowing seven steps (Planck Collaboration 2015).
(i) CMB cleaning: the 143 GHz Planck-HFI map is extrapo-lated to the other bands according to a CMB spectrum and removed from the maps at other frequencies.
(ii) Galactic cirrus cleaning: the IRAS/IRIS 100 µm map, con-sidered as a “warm template” of Galactic cirrus, is extrapo-lated to the Planck-HFI bands, taking into account the local colour, and removed from the maps following the prescrip-tion of the CoCoCoDeT algorithm (Montier et al. 2010). (iii) Construction of excess maps: the 545 GHz excess map is
defined as the difference between the cleaned 545 GHz map and a power law interpolated between the cleaned maps at 857 GHz and 353 GHz.
(iv) Point source detection in the excess maps: we
apply a Mexican hat type detection algorithm
(Gonz˜alez-Nuevo et al. 2006) with a size parameter of R = 5′in the 545 GHz excess maps.
(v) Single frequency detection: we also require a detection in each cleaned map at 857 and 353 GHz.
(vi) Colour-colour selection: we apply two criteria on the S545/S857(i.e., 545 GHz to 857 GHz) and S353/S545(i.e.,
353 GHz to 545 GHz) flux density ratios to select the red-dest sources.
This produces a dozen hundred candidates on the cleanest 26 % region of the submm sky. We note that this procedure has been set-up early on in the Planck project. The final catalog of Planck high-z candidates is being generated using a similar (but not exactly identical) method which will be described in a forthcoming paper (Planck Collaboration 2015). In particular, the CMB estimate will not be the 143 GHz HFI map, but instead the CMB derived by component separation. The present paper focuses on the first candidates followed up by Herschel.
2.1.2.Planck PCCS sources
We make use of the PCCS to choose a sample of high-z sources selected by the expected peak in their thermal dust spectrum in the rest-frame far infrared. Our four step proce-dure here is based on the work of Negrello et al. (2010). First of all, we use the 857 GHz Galactic mask, keeping 52 % of
3 Planck data (maps and catalogues) can be
Figure 1. SPIRE maps at 250, 350, and 500 µm of one typical overdensity in our sample (a Planck cold source of the CIB – Cosmic
Infrared Background). The thick white contour at 350 µm shows the Planck IN region at 857 GHz, and the same is true for the map at 500 µm at 545 GHz (i.e., the 50 % contour of the Planck source maximum). Thin contours correspond to the overdensity of SPIRE sources at each wavelength, marked at 2 σ, 3 σ, etc. (see Sect. 4.2 for details). SPIRE identifies a few (typically 5 to 10) sources inside the Planck contour. We use these contours to separate the IN and OUT zones. These data come from a 7′×7′SPIRE scan
(see Table 1 and Sect. 2.3).
Figure 2. SPIRE maps at 250, 350, and 500 µm for a lensed source. In this case a single very bright SPIRE source is detected. This
source is confirmed at the Jy level and has a redshift of z = 3.0, measured from multi-line spectroscopy acquired at IRAM using EMIR (Canameras 2015). The white contour is the Planck IN region, defined in the same way as in Fig. 1. These data come from a 7′×7′SPIRE scan (see Table 1 and Sect. 2.3).
the sky (Planck Collaboration Int. VII 2013). Secondly, we se-lect all the sources with S/N > 4 at 545 GHz and the colours S857/S545 < 1.5 (where Sν is the flux density at frequency
ν in GHz) and S217 < S353. Thirdly, we inspect each source
with respect to the NASA/IPAC Extragalactic Database (NED), IRAS maps (Neugebauer et al. 1984), and optical maps using ALADIN. Any object identified as a local galaxy, a bright radio source or Galactic cirrus is removed (about half of the objects). Finally, we remove any PCCS source already falling in the H-ATLAS or South Pole Telescope (SPT) survey fields.
Table 1. SPIRE programmes following-up Planck cold sources
of the CIB. HPASSS is composed of 124 sources selected from the maps (two were repeated, so 126 fields), together with 28 from the PCCS (four were cirrus dominated, so 24 net fields). A total of 228 sources are useable (Sect. 2.2).
Programme No. of fields Map size AORatime OT1 . . . 10 15′×15′ 3147 s
OT2 . . . 70 7′×7′ 838 s
HPASSS . . . 124 (126) (9b) + 24 (28) 7′×7′ 838 s Total . . . 228 (234)
a Astronomical Observation Request.
b Nine HPASSS sources are from archival data.
the targeted redshift range. Our main goal was to identify the counterparts contributing to the Planck-HFI detections, hence the focus on rapid SPIRE follow-up of a maximum number of targets. Obtaining PACS observations would have enhanced the angular resolution and wavelength coverage, but for these short observations would not have resulted in further constraints in most cases.
The samples are summarized in Table 1, together with the map sizes and time per Astronomical Observation Request (AOR). The OT1 sample refers to the Herschel first call for Open Time observations in July 2010 (P.I.: L. Montier). The OT2 sample refers to the Herschel second call for Open Time observations in September 2011 (P.I.: H. Dole). Figs. 1 and 2 show examples of this observation strategy. The Herschel and Planck All-Sky Source Snapshot Legacy Survey (HPASSS) was set up by the Planck collaboration (P.I.: H. Dole) in the response to ’must-do’ DDT programme in June 2012. At that time, the preliminary PCCS was internally released to the collaboration, and we could benefit from the unique and timely Herschel and Planck synergy when the Planck products were approaching their final state and when Herschel surveys already demonstrated the efficiency of the SPIRE observations. We note that the SPIRE data from 9 Planck fields included in the HPASSS sample are Herschel archival data: 4C24.28-1 (program: OT2 rhuub 2); NGP v1,
NGP h1, NGP h2, SGP sm3-h, GAMA12 rn1, NGP v8,
NGP h6 (KPOT seales01 2); Lockman SWIRE offset 1-1
(SDP soliver 3); and Spider-1 – (OT1 mmiville 2). 2.3. SPIRE data processing, total SPIRE sample and
definition of IN and OUT Planck regions
The SPIRE data were processed starting with ’Level 0’ using
HIPE10.0 (Ott 2010) and the calibration tree ’spire cal 10 1’ using the Destriper module4https://nhscsci.ipac.caltech.edu as
mapmaker (baselines are removed thanks to an optimum fit be-tween all timelines) for most observations. For 33 AORs, we had also to remove some particularly noisy detectors (PSWB5, PSWF8, and PSWE9 affected for Operational Days 1304 and 1305) from the Level 1 timelines. In that case, we processed the data with the naive scan mapper using a median baseline removal (the destriper module worked with all bolometers in Level 1). Turnarounds have been taken into account in the processed data. The useable sky surface area is thus extended, and goes beyond the nominal 7′×7′or 15′×15′as specified in the AORs (Table 1).
4 Scan map destriper details: https://nhscsci.ipac.caltech.edu/sc/index.php/Spire/PhotScanMapDestriper
Table 2. SPIRE 1 σ total noise (instrument + confusion) levels
measured in various fields, in mJy. Our Planck fields are denoted ’Planck high-z’. 1 σ noise level Field 250 µm 350 µm 500 µm Planck high-z . . . . 9.9 9.3 10.7 HerMES Lockman-SWIRE . . . 10.1 10.5 12.0 HLS . . . 14.1 12.6 14.2
Since we are close to the confusion limit, we included a check to confirm that the non-uniform coverage imposed by including the edges does not change the source detection statistics.
We have a total of 234 SPIRE targets. Of these, two fields were repeated observations, and have been used to check the ro-bustness of our detections. Four fields from the PCCS appear as cirrus structures: diffuse, extended submillimetre emission with-out noticeable point sources. These fields were removed from the sample. This means that four out of 234 fields (1.7 %) were contaminated by Galactic cirrus. The success rate of avoiding Galactic cirrus features is thus larger than 98 %, thanks to our careful selection on the Planck maps of the cleanest 35 % of the sky.
Our final sample thus contains 228 fields (i.e., 234 minus two repeated fields minus four cirrus-dominated fields). They are composed of (See Table 1): 10 sources from OT1; 70 sources from OT2; 124 objects from HPASSS CIB; and 24 sources from HPASSS PCCS.
Each SPIRE field of a Planck target is then separated into two zones: IN and OUT of the Planck source at 545 GHz (the frequency where our selection brings the best S/N ratio). The IN region boundary is defined as the 50 % Planck intensity contour, i.e., the isocontour corresponding to 50 % of the peak Planck flux, and encompasses the Planck-HFI beam. The OUT region is defined to be outside this region (see the thick white contours in Figs. 1 and 2).
2.4. Ancillary SPIRE data sets
For a first characterization of our sample, and to infer whether it is different from other samples of distant galaxies observed with SPIRE, we will compare with the data obtained in other SPIRE programmes. Since we suspect that most of our targets contain overdensities like proto-clusters of galaxies, it will be useful to contrast it with samples of galaxy clusters at lower redshift, as well as against blank fields. Our two comparison samples are as follows.
1. The HerMES5 (Oliver et al. 2010) and H-ATLAS6
(Eales et al. 2010) public data as reference fields. In particular, we will be using the ’level 5’ Lockman HerMES field, which has a similar depth to our SPIRE observations. 2. The SPIRE snapshot programme of local or massive
galaxy clusters (Egami et al. 2010) including: the “Herschel Lensing Survey” (HLS); the “SPIRE Snapshot Survey of Massive Galaxy Clusters” (OT1); and the “SPIRE Snapshot
Figure 3. Photometric accuracy from Monte Carlo source injections in real maps at 250, 350 and 500 µm. The 3 σ (dotted line) and
4 σ (dashed line) levels are also shown. See Sect. 3.2 for details.
Figure 4. Completeness levels from Monte Carlo source injections in SPIRE maps of different data sets at 250, 350 and 500 µm.
Plotted are our Planck fields (black), HerMES sources (orange), and Herschel Lensing Survey (HLS) clusters (blue). The 50 % (dotted line) and 80 % (dashed line) completeness levels are also shown. See Sect. 3.2 for details.
Survey II, Using SPT/CODEX Massive Clusters as Powerful Gravitational Lenses” (OT2)7.
Finally, we use the OT2 data from “The highest redshift strongly lensed dusty star-forming galaxies” (P.I. J. Vieira) pro-gramme to carry out technical checks, since those AORs are very similar to ours.
3. SPIRE Photometry
3.1. Photometric analysis
We first measure the noise in each map at each wavelength (channel) by fitting a Gaussian function to the histogram of the maps and deriving the standard deviation, σchannel, a mixture of
confusion noise (Dole et al. 2003; Nguyen et al. 2010) and de-tector noise. The values are reported in Table 2. Data are sky sub-tracted and at mean 0, as done in HIPE 10. All pixels are used to plot the noise histogram. Histogram have a Gaussian shape, with a bright pixel tail. The bright pixel tail contribution (larger than 3
7 Herschel public data can be
down-loaded from the Herschel Science Archive: http://herschel.esac.esa.int/Science_Archive.shtml
sigma) to the histogram is nearly 1% in number for each field at 350um. Then we extract the sources using a blind method and by a band-merging procedure described in the following sections.
3.2. Blind catalogues
We detect blindly, i.e., independently at each frequency, the sources using StarFinder (Diolaiti et al. 2000). We use Gaussian point spread functions (PSFs) with FWHM of 18.15′′ for 250 µm, 25.15′′ for 350 µm, and 36.3′′ for 500 µm8. These
individual band catalogues are used for checking photometric accuracy and completeness, and to produce number count esti-mates. The catalogues are then band-merged in order to quantify the colours of the sources.
We employ Monte Carlo simulations to check our photom-etry at each wavelength: injection of sources in the data, and blind extraction, in order to measure the photometric accuracy (Fig. 3). As expected, the photometric accuracy is of the order of 10 % at flux densities larger than a few tens of mJy, and de-creases towards smaller flux densities closer to our noise level
Figure 5. Differential Euclidean-normalized number counts, S2.5dN/dS , for various data sets (not corrected for flux-boosting or incompleteness) used to measure the relative behaviour. Symbols show: Planck IN region (red circles, 228 fields); Planck OUT regions (black circles); Planck whole field, i.e., IN+OUT (blue circles); sample of 535 z < 1 clusters of the HLS (open square); and HerMES Lockman SWIRE blank field (crosses). Our SPIRE sources corresponding to the Planck IN regions clearly show an excess in number density at 350 and 500 µm, illustrating their red colour, and a potentially higher redshift than average. See Sect. 4.1 for details.
(dominated by confusion) at around 30 mJy. The completeness levels are also measured, and reported in Fig. 4 and Table 3.
3.3. Band-merged catalogues
We use the 350 µm map-based catalogue from StarFinder as the input catalogue. This wavelength has two advantages: first, its angular resolution meets our need to identify the sources; and second, it is consistent with the Planck colour selection, which avoids the many low-z galaxies peaking at 250 µm. As will be shown in Sect. 4.1, our sample has an excess in num-ber at 350 µm compared to reference samples drawn from the HerMES Lockman SWIRE level 5 field or the low-redshift HLS cluster fields, in agreement with the existence of an overdensity of SPIRE sources in the Planck beam.
Our band-merging procedure has three steps. First of all, we optimize the measurement of the source position using the 250 µm channel, where available. Secondly, we measure a preliminary flux density on each source, which will serve as a prior to avoid unrealistic flux measurements while deblending. Thirdly, we perform spatially-simultaneous PSF-fitting and deblending at the best measured positions with the newly determined prior flux densities as inputs for
FastPhot (Bethermin et al. 2010b). The details are described in Appendix A. This method provides better matched statis-tics (more than 90 % identifications) than the blind extraction method performed independently at each wavelength (Sect. 3.2, Table A.2 vs. Table A.1). We finally have a total of about 7100 SPIRE sources, of which about 2200 are located in the Planck IN regions (giving an average of about 10 SPIRE sources per Planck IN field).
4. Statistical Analysis of the Sample
4.1. Number counts: significant excess of red sources We compute the differential Euclidean-normalized number counts S2.5dN/dS , with S being the flux density at wavelength λ, and N the number of sources per steradian. The counts are not corrected here for incompleteness, or for flux
boost-Table 3. 80 % completeness levels obtained from Monte Carlo
source injection in the SPIRE maps for different fields, in mJy. Our Planck fields are denoted “Planck high-z”.
80 % Completeness
Field 250 µm 350 µm 500 µm
Planck high-z . . . . 35.2 37.0 40.7 HerMES Lockman-SWIRE . . . 35.0 38.4 42.7 HLS . . . 49.3 48.5 54.3
ing (Eddington bias), since we are interested in the relative be-haviour between the samples. The detected sources used here are extracted using the blind technique (Sect. 3.2). We cut the sam-ples at 4 σ for these counts (σ values in Table 2). We used five different data sets to estimate the number counts:
• Planck IN+OUT, our 228 Planck entire fields, each covering about 20′×20′;
• Planck IN, our 228 Planck fields, using only the central parts corresponding to the Planck 50 % contour level (determined separately for each source), called the IN region;
• Planck OUT, our 228 Planck fields, using only the part exte-rior to the Planck 50 % level, called the OUT region;
• HerMES Lockman SWIRE, HerMES level 5 field in
Lockman (Oliver et al. 2010), covering 18.2 deg2;
• HLS, 535 cluster-fields of Egami et al. (2010), five from the Herschel Lens Survey KPOT, 282 from OT1, and 248 from OT2, hereafter refered to as HLS.
The number counts of all these data sets are plotted in Fig. 5 and reported in Appendix B in Tables B.1, B.2, and B.3. We derive from the total number count the following results.
300 mJy, where the numbers are small. This means that, on average, those fields do not show any strong deviations between them.
(ii) The observed counts (uncorrected for incompleteness) show the characteristic shape of the Eddington bias: a cut-off below about 40 mJy, an excess around 50 mJy, and a behaviour compatible with the models at higher flux den-sities. It is beyond the scope of this paper to re-derive un-biased number counts (e.g., Glenn et al. 2010; Oliver et al. 2010; Clements et al. 2010), as we only focus on the rela-tive trends.
(iii) The IN counts at large flux densities (S > 300 mJy) show a systematic excess at 350 and 500 µm over the counts in the wide blank field (Lockman) as well as cluster fields. At 300 mJy, the overdensity factors are, respectively, 4.1 and 3.7 at 350 µm, and about 20 and 16 at 500 µm. This chro-matic excess (i.e., larger excess at longer wavelengths) is consistent with the hypothesis of the presence of a popula-tion of high-redshift lensed candidates at z = 2–4. We will show in Sect. 5 that this is indeed the case.
(iv) The IN counts at 250 µm are a bit higher (for S250 <
100 mJy) than in Lockman, and are lower at larger flux densities. This means that the SPIRE counterparts of the Planck sources have, on average, counts that deviate little from blank fields at this wavelength.
(v) The IN counts at 350 µm for S350≃50 mJy are a factor
be-tween 1.9 and 3.4 higher than in Lockman and in the clus-ter fields. This means that, on average, the SPIRE coun-terparts of the Planck sources have a significant excess of 350 µm sources compared to wide blank fields or z < 1 HLS cluster fields. This is expected, given our Planck se-lection criteria.
(vi) The IN counts at 500 µm for S500≃50 mJy are a factor of
2.7–8 higher than in the Lockman and cluster fields. As in item (v) above, this means that, on average, the SPIRE im-ages of the Planck source targets show a significant excess of 500 µm sources compared to wide blank field or z < 1 HLS cluster fields.
(vii) The OUT counts are compatible with the wide blank extra-galactic fields, as well as the cluster fields. Our OUT zones can thus also be used as a proxy for the same statistics in blank fields.
As a conclusion, the SPIRE observations of the Planck fields reveal “red” sources. The SPIRE images exhibit a significant ex-cess of 350 and 500 µm sources in number density compared with wide blank fields (HerMES Lockman SWIRE of the same depth) or fields targeting z < 1 galaxy clusters (HLS). This sig-nificant excess should be expected given the Planck colour se-lection, and is now demonstrated with secure SPIRE detections. It is therefore clear that there is no significant contamination by cirrus confusion in our Planck sample, and that where there is a Planck high-z candidate, Herschel detects galaxies.
4.2. Overdensities
We compute the dimensionless overdensity contrast δλ of our
fields at wavelength λ via
δλ=
ρIN−ρOUT
ρOUT
, (1)
where ρINis the surface density of SPIRE sources in the Planck
IN region, and ρOUT is the mean surface density of SPIRE
sources computed in the Planck OUT region, at SPIRE wave-length λ. We have already shown (Sect. 4.1 and Fig. 5) that the OUT region has a density equivalent to that of blind surveys, and is thus a good estimate of ¯ρ, the mean surface density. To reduce the Poisson noise, we use the counts from all the OUT regions.
The overdensity contrasts δλ extend up to 10, 12, and 50
at 250, 350, and 500 µm, respectively, with a median overden-sity δλ of δ250 = 0.9, δ350 = 2.1, and δ500 = 5.0. This means
that our Planck IN regions have an excess of SPIRE sources. Indeed, there are 50 fields with δ500>10, and 129 with δ500>4
(with significance levels always higher than 4 σδ, see below). At
350 µm, there are 19 fields with δ350>5, 37 fields with δ350>4
and 59 fields with δ350 > 3, as shown in Fig. 6 (left panel, in
blue). In Appendix D, we also use the densities measured with AKDE to estimate the overdensities.
How significant are these overdensity contrasts? To quantify this we compute the mean density field using the AKDE algo-rithm (adaptative kernel density estimator, see Valtchanov et al. 2013 and also Pisani 1996; Ferdosi et al. 2011). The principle is to generate a two-dimensional density field based on the po-sitions of the sources from a catalogue, filtered (smoothed) ac-cording to the source surface density. From this smoothed field, we compute the standard deviation and hence we derive the sig-nificance σδof the overdensity. We also run 1000 Monte Carlo
runs to get a better estimate of the scatter on σδ (by creating
AKDE density maps using random source positions – but oth-erwise using the real catalogs of each field – and measuring the RMS over those 228000 realizations). All our fields show overdensities larger than 1.8 σδ, with a median of 7 σδ (Fig. 6
right panel). The typical number of sources in an overdensity is around 10, but with a fairly wide scatter.
We can then choose to select the reddest sources, this be-ing a possible signature of a higher redshift or a colder dust temperature. We define the SPIRE red sources with the follow-ing cuts in colour based on the distributions shown in Fig. 7: S350/S250>0.7; and S500/S350>0.6. When selecting only the
red sources, the overdensity significance increases: The mean is now 12 σδand the median 9 σδ. We have 50 % of the sample at
10 σδor more (when selecting the red sources), which is more
than a factor of 3 larger than when selecting all SPIRE sources. 23% of the sample is above 15 σδ (Fig. 6 right panel), i.e., 51
fields.
These high significance levels can be contrasted with the mean δ500 ≃ 0.25 obtained by Rigby et al. (2014) at the
loca-tions of 26 known protoclusters around very powerful radio-galaxies, drawn from the list of 178 radiogalaxies at z > 2 of Miley & de Breuck (2008). They can also be compared to the few similar examples in Clements et al. (2014), with at most 4.7 σδat 350 µm. We show in Fig. 6 right panel, in orange, the
cumulative normalized significance σδof 500 random positions
in the Lockman field; Using this test sample illustrates the high significance of the overdensities of our sample.
Examples of some overdensities are shown in Appendix E, where we present a gallery of representative SPIRE data. 4.3. Colours of the sources
Using the band-merged catalogue (Sect. 3.3), we can derive the colours of the sources, i.e., the ratios of the observed flux densi-ties S500/S350vs. S250/S350. In this colour-colour space, redder
sources will have higher S500/S350ratios and lower S250/S350
−2 0 2 4 6 8 10 12 overdensity contrast δ350 0 50 100 150 200 250 cu mu lat ive nu mb er of so urc es overdensity all δ350 0 10 20 30 40 50 significance σδ 0.0 0.2 0.4 0.6 0.8 1.0 no rm ali ze d c um ula tiv e n um be r o f so urc es
signif. σδ Planck/Herschel red sources
signif. σδ Planck/Herschel all sources
signif. σδ Lockman sources
Figure 6. Left: Cumulative histogram of the overdensity contrast δ350(blue) of each Planck field (based on 350 µm SPIRE sources).
Overdensities are fairly large, with 59 fields having δ350>3. Right: Cumulative (normalized) statistical significance in σ derived
from the density maps. Blue represents all our SPIRE sources, red represents only redder SPIRE sources, defined by S350/S250>0.7
and S500/S350 >0.6, and orange 500 random fields in Lockman. Most of our fields have a significance greater than 4 σ, and the
significance is higher still for the redder sources. See Sect. 4.2 for details.
Figure 7. Colour counts: source surface density as a function of the SPIRE colour, S350/S250(left), and S500/S350(right). Histograms
are: red solid line, Planck IN; black line, Planck OUT; green, Planck IN lensed fields only; blue line, z < 1 HLS clusters; and orange dashes, Lockman SWIRE. The Planck IN sources (total and/or lensed sources) show a much higher surface density than other samples, owing mainly to our all-sky search strategy. See Sect. 4.3 for details.
IN; black, Planck OUT; green, only the lensed fields; blue, HLS; and dashes, HerMES). The IN sources show three times larger surface densites in S350/S250, and four times larger surface
den-sities in S500/S350 than OUT and Lockman sources. The IN
source distribution peaks at much higher surface density than any other sample, suggesting that our sample is dominated by red and overdense SPIRE sources.
Following the approach of Amblard et al. (2010), we gener-ate the SED of 106 modified blackbodies with 10 % Gaussian noise and explore three parameters: the blackbody temperature (T ) in the range 10–60 K; the emissivity (β) in the range 0–2;
and the redshift for z = 0–5. For each set of parameters we convert the fixed, observed SPIRE wavelengths into rest frame wavelength (at redshift z, thus varying with z) using: λrest =
λSPIRE/(1 + z). We then calculate the flux at each wavelength
using λrestand compute the colours S250/S350and S500/S350for
Figure 8. Herschel-SPIRE colour-colour diagram of IN sources (black dots) on top of redshifted blackbodies. Left: colour codes
redshift. Right: colour codes dust temperature in kelvin. Although there is a degeneracy between redshift and temperature, the colours suggest a range of z ∼ 1.5–3 for the SPIRE sources corresponding to the Planck IN regions. The upper right symbol gives the typical (median) error bars of the measured points. See Sect. 4.3 for details.
Figure 9. Classification of 228 Planck fields with
Herschel-SPIRE. The x-axis represents the maximum flux of the SPIRE source within the IN region: lensed source candidates are se-lected if they fall above 400 mJy (i.e., the right column, seven sources). The y-axis represents the wavelength at which the brightest SPIRE source peaks in the IN region. Overdensities are thus selected in the lower left four cells. The colour represents the number of fields in each cell (as shown in the colour bar). the numbers in the cells are the number of fields and percent-age in each cell. Our sample is thus dominated by overdensities peaking at 350 µm. See Sect. 4.4 for details.
this suggestive result (Amblard et al. 2010; Pope & Chary 2010; Greve et al. 2012), and will be discussed in Sect. 6.3.
Figure 10. A high-z cluster candidate observed by Planck, Herschel, and Spitzer-IRAC (image covering 0.5′×0.3′). We show the IRAC channel 1 (3.6 µm) image, with SPIRE 350 µm white contours overlaid. Colour contours represent statistical significance of the local overdensity, from light blue to red: 3, 4, 5, 6, and 7 σδ. See Sect. 6.1 for details.
4.4. Classification of the sources
We sort the fields using two criteria: (1) the SPIRE chan-nel at which the source has its maximum flux density; and (2) its flux density at 350 µm. We will use this two-parameter space to classify the sources. This classification provides the first set of information about the sources, although a definitive classifi-cation will require follow-up data to confirm the nature of the sources.
In this space, the lensed candidates will populate the “red” and “bright” areas (typically 350 or 500 µm peakers, and Sν >
400 mJy). This does not necessarily mean that all the sources in this area will be strongly gravitationally lensed (although this is confirmed by further follow-up, see following sections).
The overdensity candidates will be located in a different area of this two-parameter space than the lensed candidates. They will tend to populate the fainter end (typically Sν < 200 mJy)
and will have red colours (typically 350 or 500 µm peakers). We do not expect many 250 µm peakers because of the Planck colour selection, for which there is a bias towards red-der colours and thus potentially higher-redshift sources.
Fig. 9 summarizes the classes of SPIRE identifications of the Planck high-z (meaning z > 1.5) candidates, based on those criteria. We show that the vast majority of our sample is com-posed of overdensities of 350 µm peakers. The number of lensed candidates is rather small in comparison, amounting to seven sources based on this criterion. All are confirmed to be strongly gravitationally lensed galaxies (Canameras 2015). We will show later that visual inspection led us to discover a few more lensed sources.
The dashed lines in Fig. 7 show the colour vs. surface den-sity of the lensed candidate population, whose red colours are confirmed.
Finally, we derived number counts (as in Fig. 5) but now by separating the overdensities from the lensed sources that dom-inate the counts at large flux densities, typically S > 250 mJy (shown in Appendix C in Fig. C.1). This separation suggests that the excess at large flux densities in the number counts of our sample is due to the presence of bright lensed sources, compared to reference samples (HLS, HerMES).
5. Strongly gravitationally lensed source candidates
5.1. Validation of existing lensed sources
Negrello et al. (2007) and Bethermin et al. (2012) predicted that a small, but significant fraction of very bright high-redshift (z > 2) submillimetre galaxies are strongly grav-itationally lensed, dusty starbursts in the early Universe. Negrello et al. (2010) presented observational evidence of these predictions. In addition, Fu et al. (2012) confirmed the na-ture of the source H-ATLAS J114637.9-001132 as a strongly gravitationally lensed galaxy at z = 3.3; this source was part of the first release of the Planck ERCSC Catalogue (Planck Collaboration VII 2011; Planck Collaboration 2011), and fell fortuitously into the H-ATLAS survey field (see also Herranz et al. 2013). Another independent confirmation that Planck sources can be strongly gravitationally lensed came from the source HLS J091828.6+514223, confirmed as a bright, z = 5.2 gravitationally lensed galaxy behind the massive intermediate-redshift galaxy cluster Abell 773 (Combes et al. 2012), as part of the Herschel Lensing Survey (Egami et al. 2010). This source was independently found in our survey and is part of the Herschel/SPIRE OT2 selection. It is excluded from our sample to satisfy the condition of non-redundancy.
5.2. Previously unknown gravitationally lensed sources In the absence of extensive follow-up, it is challenging to distin-guish between single or multiple strongly gravitationally lensed galaxies behind the same foreground structure and overdensi-ties of intrinsically bright submillimetre galaxies, for all but the brightest gravitationally lensed sources. Moreover, at flux den-sities of about S350= 250 mJy and above, isolated SPIRE point
sources may turn out to be associations of multiple FIR galaxies when observed at higher spatial resolution (Ivison et al. 2013).
As a first step towards identifying the gravitational lensed candidates in our sample (Sect. 4.4), we therefore focused on those targets where SPIRE shows only a single, very bright source, with the typical FIR colours of high-redshift (z > 2) galaxies associated within one Planck beam. We thus identified seven isolated SPIRE point sources, as described in Sect. 4.4, plus five others at slightly fainter flux densities. All have peak flux densities at 350 µm, including 11 with S350 = 300–
1120 mJy (Canameras 2015). Six of these galaxies were taken from the PCCS, the remaining six originate from the sample of Planck Collaboration (2015). Although the initial selection of lensed candidates was carried out by eye upon the reception of the SPIRE imaging, seven of these targets were also identified with our automatic classification (Sect. 4.4). We used the IRAM 30-m telescope to obtain firm spectroscopic redshifts via a blind CO line survey with the wide-band receiver EMIR. We iden-tify 2−6 lines per source, which confirms they are at redshifts z = 2.2–3.6. Interferometry obtained with the IRAM Plateau de Bure interferometer and the Submillimeter Array, as well as several empirical calibrations based on FIR luminosity and dust temperature (following Harris et al. 2012), and CO line lu-minosity and line width (following Greve et al. 2012), demon-strate that these are indeed strongly gravitationally lensed galax-ies, amongst the brightest on the submm sky (Canameras 2015).
6. Candidate high-zoverdensities
Without rejecting the possibility that the observed overdensities of red SPIRE sources could be chance alignments of structures giving coherent colours (e.g., Chiang et al. 2013), we can pon-der the nature of those overdensities. Could they be high-redshift intensively star-forming galaxy proto-clusters? Indeed, recent studies, such as Gobat et al. (2011), Santos et al. (2011, 2013, 2014), and Clements et al. (2014) confirm the presence of high redshift (z > 1.5) galaxy clusters emitting enough energy in the submillimetre to be detected. Those previously detected clusters are, however, in a different state of evolution than our candidates. Many of the confirmed clusters exhibit X-ray emission, suggest-ing already mature and massive clusters. The followsuggest-ing sections investigate different aspects of our sample, which is uniquely se-lected by strong submm emission over the whole sky, enabling us to unveil a rare population.
6.1. First confirmations
It is beyond the scope of this paper to summarize all the follow-up observations conducted so far, but we give here three high-lights, which will be discussed in more detail in subsequent pa-pers.
Figure 11. Left: SPIRE 3-colour image of one Planck high-z overdensity candidate (located in the centre), in an image covering
about 20′×20′. We note the bluer source on the left (circled), which corresponds to the galaxy cluster shown on the right. Right:
a confirmed z = 1.58 galaxy cluster XMMU J0044.0-2033 (Santos et al. 2011) which lies close to our Planck high-z source. The background image is a two colour composite of Spitzer-IRAC channel 1 (3.6 µm) and channel 2 (4.5 µm), covering about 3′×2′. The contours are Herschel-SPIRE 350 µm. See Sect. 6.1 for details.
Figure 12. Stacks of SPIRE 350 µm data (8.7′×8.7′): (a) 220 Planck Herschel fields; (b) 278 HLS clusters; (c) 500 sources in the
1.7 and 2.0, we find an average dust temperature for the sources of, respectively, 27K and 35K.
0 1 2 3 4 5 Redshift [z] 0 50 100 150 200 250 300 350 400 450 # of so ur ce s 25K 30K 35K 40K 45K
Figure 13. SPIRE photometric redshift distribution of the
roughly 2200 SPIRE sources in the IN regions, as a function of the fixed assumption for dust temperature: Td = 25, 30, 35,
40, and 45 K (from left to right). See Sect. 6.3 for details.
Another field has been imaged by Spitzer as part of our GO-9 programme – see Fig. 10 for the IRAC 3.6 µm image with the SPIRE 350 µm contour. The colour ratios of IRAC flux densi-ties 3.6 µm/4.5 µm of the sources detected in the overdensity exhibit red colours, indicating their probable high-redshift na-ture (Papovich 2008). The overdensity of IRAC 3.6 µm sources has a statistical significance of 7 σδ (Martinache et al., 2014,
in prep.). While not yet confirmed as a proto-cluster or cluster through spectroscopy, this structure, seen with Spitzer at near-infrared wavelengths, supports the hypothesis of strongly clus-tered highly star-forming infrared sources.
Finally, we have also found one SPIRE field with overlap-ping data from the literature. The observations from Santos et al. (2011) confirm the presence of a z = 1.58 cluster using op-tical/NIR spectroscopy. This particular high-z cluster has also been detected in X-rays, leading to a total mass estimate of about 3–5 × 1014M⊙. It is located a few arcminutes from one of our
targets (Fig. 11 right), so is not formally an identification of our Planck source, but rather a coincidence. While not connected to our target source here, this cluster can nevertheless bring use-ful information. The SPIRE colours of this cluster are bluer than our sources. Since our target shows redder SPIRE colours, this suggests it has a higher redshift or a cooler dust temperature.
6.2. Large clustering of our sample revealed with stacking We can investigate if the observed overdensities have differ-ent clustering properties than two test samples: (1) the HLS massive z < 1 clusters; and (2) 350 µm peaker sources in the HerMES Lockman field brighter than 50 mJy. To do so, we use a stacking analysis (see, e.g., Montier & Giard 2005; Dole et al. 2006; Braglia et al. 2011), here applying the method of Bethermin et al. (2010a). We stack the following SPIRE 350 µm data: (a) the 220 overdensities of our sample; (b) 278 HLS clusters; (c) 500 bright 350 µm peaker sources in the
Lockman field; and (d) 500 random positions in the HerMES Lockman field as a null test. The stacks are presented in Fig. 12. We can see that the Planck fields show clear and signifi-cant extended (a few arcminutes) emission due to the cluster-ing of bright submm sources. This kind of extended emission of clustered submm sources is not observed in the HLS clus-ters, nor around 350 µm peaking HerMES sources. The ran-dom stack validates the absence of a systematic effect in stack-ing (Dole et al. 2006; Bethermin et al. 2010a; Viero et al. 2013; Planck Collaboration XVIII 2014). This proves that our sources are of a different nature than mature HLS clusters or average HerMES submillimetre sources.
We overplot in Fig. 12 the contours in density (see Sect. 4.2) of the SPIRE red sources (defined in Sect. 4.2). The stacks indi-cate that the Planck sample exhibits strong overdensities of red SPIRE sources. By contrast, the HLS sample shows a smooth and weak density of red sources (except for the presence of some point sources – the background lensed galaxies). As expected, the HerMES sample shows a stack consistent with the SPIRE PSF, while the random sample is consistent with noise.
6.3. Dust temperatures, photometric redshifts, luminosities, and star formation rates
With the hypothesis that our overdensities are actually gravi-tationally bound structures, i.e., clusters of star-forming, dusty galaxies, the overdensity colours (Fig. 7) suggest a peak redshift around z = 1.5–3 (Fig. 8). Using a modified blackbody fit (with
β =1.5) and fixing the dust temperatures to Td= 25, 30, 35, 40,
and 45 K, we obtain the redshift distributions shown in Fig. 13 for the roughly 2200 sources found in the IN regions for these overdensities. For Td= 35 K, the distribution peaks at z = 2. For
higher dust temperatures, the peak of the distribution is shifted towards higher redshifts, since dust temperature and redshift are degenerate for a modified blackbody, where Td/(1+z) =constant.
Many studies (e.g., Magdis et al. 2010; Elbaz et al. 2011; Greve et al. 2012; Magdis et al. 2012; Symeonidis et al. 2013; Weiss et al. 2013; Magnelli et al. 2014) suggest that dust tem-peratures of z ∼ 2 sources are typically of the order of 35 K, consistent with the few measurements in hand (see Sect. 6.1). With the conservative assumption that Td = 35 K for all sources
(e.g., Greve et al. 2012), we can derive the IR luminosity for each SPIRE source (by integrating between 8 µm and 1 mm), and find that it has a broad distribution peaking at 4 × 1012L
⊙
(Fig. 14). Assuming that the conventionally assumed relation-ship between IR luminosity and star formation (Kennicutt 1998; Bell 2003) holds, and that AGN do not dominate (as was also found for the objects previously discussed by Santos et al. 2014 and Clements et al. 2014), this would translate into a peak SFR of 700 M⊙yr−1 per SPIRE source (Fig. 15). If we were to
use colder dust temperatures, the peak SFR would be at about 200 M⊙yr−1 per SPIRE source (Fig. 15). If we were to use
warmer dust temperatures, the SFR would increase because the implied redshift would be higher and still compatible with other observations, e.g., the compilation of submillimetre galaxies in Greve et al. (2012), where LIRis measured in the range 3 × 1012–
1014L
⊙and Tdin the range 30–100 K, for redshifts above 2.
of 4 × 1013L
⊙. This translates to a peak SFR of 7 × 103M⊙yr−1
per structure, as seen in Fig. 16.
Our estimates of LIR are within the range of FIR
luminosi-ties expected for massive, vigorously star-forming high-z struc-tures (e.g., Brodwin et al. 2013), which are perhaps protoclus-ters in their intense star formation phase, and are consistent with the four bound structures found in Clements et al. (2014). These z > 2 intensively star forming proto-clusters are also ex-pected in some models, e.g., the proto-spheroids of Cai et al. (2013), or the halos harbouring intense star formation discussed in Bethermin et al. (2013).
0.0 0.2 IR Luminosity per source [L0.4 0.6 0.8 1.0
⊙] 1e13 0 100 200 300 400 500 600 No . o f so urc es 25K 35K
Figure 14. Infrared luminosities (LIR) of the approximately 2200
SPIRE sources in the IN regions, as a function of the fixed dust temperature: 35 K (purple, right histogram), and also shown for illustration 25 K (blue, left histogram). See Sect. 6.3 for details.
6.4. Massive galaxy clusters in formation?
It is of course tempting to postulate that all (or at least a subset) of our sources include massive galaxy clusters in the process of formation. However, demonstrating this conclusively with the present Herschel photometry alone is not possible. We have therefore undertaken a comprehensive multi-wavelength photo-metric and spectroscopic follow-up campaign to directly con-strain the nature of our sources and in particular, to investigate whether they are good candidates for being the progenitors of to-day’s massive galaxy clusters seen during their most rapid phase of baryon cooling and star formation.
Even without such an explicit analysis of the astrophysical nature of our candidates, we can already state that: (1) our esti-mated luminosities are consistent with the expected star forma-tion properties of massive galaxy clusters, as obtained from stel-lar archaeology in galaxies in nearby clusters and out to redshifts z ∼ 1; and (2) our total number of overdensities are consistent with the expected numbers of massive high-z galaxy clusters, as obtained from models of dark matter halo structure formation.
For point (1), a broad consensus has now been developed whereby most of the stellar mass in massive cluster galaxies was already in place by z ≃ 1. The most stringent observational constraint is perhaps the tight red sequence of cluster galaxies in colour-magnitude diagrams of massive galaxy clusters in the optical and near-infrared (e.g., Renzini 2006), which suggests
that massive cluster galaxies formed most of their stellar popula-tions within a short timescale of ≤ 1 Gyr, i.e., within one cluster dynamical time, with little star formation thereafter. Rest-frame optical studies of distant clusters suggest that the bright end of the red sequence was already in place by z ≃ 1 (Rudnick et al. 2009), with a likely onset amongst the most massive galaxies by
≃2, or even before (Kodama et al. 2007).
0 500 1000 1500 2000 SFR per source [M⊙ yr−1] 0 100 200 300 400 500 600 No . o f so ur ce s 25K 35K
Figure 15. Star formation rate (SFR) for individual SPIRE
sources in the overdensity sample (i.e., in the Planck IN region). For Td = 35 K, a dust temperature favoured by the literature,
(purple, right histogram) the distribution peaks at 700 M⊙yr−1.
We also show for illustration Td = 25 K (blue left histogram
peaking at 200 M⊙yr−1). See Sect. 6.3 for details.
The shortness of the star formation period in nearby clus-ters allows us to derive an order-of-magnitude estimate of the total star formation rates during this period, as implied by the fossil constraints, and the resulting FIR fluxes. To obtain upper limits on the stellar mass formed during this epoch, we use stel-lar mass estimates obtained for 93 massive X-ray-selected clus-ters with M500 = (1014–2 × 1015) M⊙at z = 0–0.6 by Lin et al.
(2012) based on WISE 3.4 µm photometry. M500is the total mass
within the central cluster regions where the mass surface den-sity exceeds the cosmological value by at least a factor of 500. Lin et al. find that in this mass range, stellar mass scales with M500as (M∗/1012M⊙) = (1.8 ± 0.1) (M500/1014M⊙)0.71±0.04. For
clusters with M500= 1 × 1014M⊙(or 2 × 1015M⊙, their highest
mass), this corresponds to M∗= 2 × 1012M⊙(or 1.5 × 1013M⊙).
We do not expect that including the intracluster light would in-crease these estimates by more than a few tens of percent (e.g., Gonzalez et al. 2007), which is a minor part of the uncertainty in our rough order-of-magnitude estimate.
These mass and timescale estimates suggest total star forma-tion rates in the progenitors of massive galaxy clusters at lower redshifts of a few ×103and up to about 2×104M⊙yr−1. Using the
conversion of Kennicutt (1998) between star formation rate and infrared luminosity (Sect. 6.3), this corresponds to LIR ≃(1013–
1014) L
⊙. This is well within the range of the global star
forma-tion rates and LFIRvalues in our sample of overdensities.
For point (2), the absence of such objects, we use the Tinker et al. (2008) halo model to compute the expected sur-face density of dark matter halos. We expect between 8500 and 1 × 107 dark matter halos at z > 2 having masses Mtot>1014M⊙
as-0 5000 10000 15000 20000 Total SFR [M⊙ yr−1] 0 10 20 30 40 50 60 70 80 90 No . o f so ur ce s 25K 35K
Figure 16. Total star formation rate (SFR) per Planck IN region,
i.e., for the overdensity sample, assuming each overdensity is an actual high-z cluster of dusty galaxies, located at a redshift given by the SPIRE photometric redshift, and using a modified blackbody fit with a fixed dust temperature of Td = 35 K (purple,
right histogram peaking at 7 × 103M⊙yr−1), this temperature
being favoured by the literature. We also show the total SFR for Td = 25 K, giving a peak at 2 × 103M⊙yr−1. The structures
could be intensively star-forming clusters. See Sects. 6.3 and 6.4 for details.
sumption that the cluster galaxies have formed most of their stellar populations within a short timescale (see above), we esti-mate that only a fraction of those halos will be observationally caught during their intense star formation phase, when they are infrared- and submm-bright. Assuming that this phase happened mainly between z = 5 and z = 1.5–2.0, we find that the upper limit on the formation period of each individual cluster, about 1 Gyr, is 3–4 times shorter than the cosmic time elapsed over this epoch, 2–3 Gyr. We would therefore expect to find a few thou-sand dark matter halos on the submm sky at any given observing epoch. This order-of-magnitude estimate is above with our find-ing of about 200 overdensities with Planck in this study, and with a few hundred to be detected in the full Planck data set by Planck Collaboration (2015), but not more than an order of mag-nitude, illustrating the need to understand the detailed processes of star formation in the most massive halos (Bethermin et al. 2013).
7. Conclusions
Our Planck sample based on a colour selection of cold sources of the CIB is overwhelmingly dominated in number by significant galaxy overdensities peaking at 350 µm, and a minority of rare, bright z > 1.5 strongly gravitationally lensed sources, among which are the brightest ever detected in the submillimetre. This confirms the efficiency of Planck to select extreme and rare submillimetre sources over the whole sky, as well as the need for higher angular resolution imaging using Herschel (and cur-rent/planned ground-based submm/mm observatories) to iden-tify and study them.
From the analysis of Herschel observations of the sample of 228 Planck cold sources of the CIB, we draw several conclu-sions.
– Less than 2 % of the fields are Galactic cirrus structures.
When a clean and controlled selection is performed on Planck, there is thus no reason to expect a large cirrus con-tamination when using a conservative Galactic mask.
– With about 93 % of the overdensities peaking at 350 µm, and
3.5 % peaking at 500 µm, our sample unambiguously selects sky areas with the largest concentrations of red SPIRE galax-ies, consisting of highly significant overdensities.
– Some of the overdensities are confirmed high-z structures,
e.g., a source at z ≃ 1.7–2.0 (Flores-Cacho et al. 2014, in prep.).
– The significance in density contrast of the overdensities (e.g.,
half of the fields above 10 σδ when selecting red SPIRE
sources) is higher than any other sample targeting proto-clusters or high-z proto-clusters in the submillimetre, confirming the relevance of the strategy of using Planck data on the cleanest parts of the sky to uncover high-z candidates.
– The SPIRE sources in the overdensity fields have a peak
redshift of z = 2 or 1.3, if we fix the dust temperature at Td= 35 K or 25 K, respectively.
– With the assumption of Td = 35 K, each SPIRE source has
an average IR luminosity of 4 × 1012L
⊙, leading to star
for-mation rates for each source peaking at 700 M⊙yr−1. If
con-firmed, these exceptional structures harbouring vigorous star formation could be proto-clusters in their starburst phase.
– Assuming the SPIRE sources are located in the same
large-scale overdensity, we derive a total IR luminosity of 4 × 1013L⊙, leading to total star formation rates of 7 ×
103M
⊙yr−1, and with around 10 detected sources per
struc-ture.
– About 3 % of our sample is composed of intensely
gravita-tionally lensed galaxies. This sample is unique, as it targets the brightest observed such sources, typically above 400 mJy at 350 µm and reaching up to the jansky level. They are all spectroscopically confirmed to lie at redshifts z = 2.2–3.6 (Canameras 2015; Nesvadba 2015).
– The novelty and efficiency of our new sample is that it
provides about 50 times more fields for a Planck-Herschel co-analysis than in existing Herschel surveys searching for serendipitous Planck sources (e.g., Clements et al. 2014), for only a fraction of the Herschel observing time.
– Our new sample exhibits high density contrasts with a high
significance: for the red sources, 30 % of our sample (about 70 fields) shows significance levels higher than 4.5 σδ, and
15 % of our sample (34 fields) are higher than 7 σδ.
